[flang] Initial debug info support for local variables. (#90905)

We need the information in the `DeclareOp` to generate debug information
for variables.  Currently, cg-rewrite removes the `DeclareOp`. As
`AddDebugInfo` runs after that, it cannot process the `DeclareOp`. My
initial plan was to make the `AddDebugInfo` pass run before the cg-rewrite
but that has few issues.
    
1. Initially I was thinking to use the memref op to carry the variable
attr. But as @tblah suggested in the #86939, it makes more sense to
carry that information on `DeclareOp`. It also makes it easy to handle
it in codegen and there is no special handling needed for arguments. For
this reason, we need to preserve the `DeclareOp` till the codegen.
    
2. Running earlier, we will miss the changes in passes that run between
cg-rewrite and codegen.
    
But not removing the DeclareOp in cg-rewrite has the issue that ShapeOp
remains and it causes errors during codegen. To solve this problem, I
convert DeclareOp to XDeclareOp in cg-rewrite instead of removing
it. This was mentioned as possible solution by @jeanPerier in
https://reviews.llvm.org/D136254
    
The conversion follows similar logic as used for other operators in that
file. The FortranAttr and CudaAttr are currently not converted but left
as TODO when the need arise.

Now `AddDebugInfo` pass can extracts information about local variables
from `XDeclareOp` and creates `DILocalVariableAttr`. These are attached
to `XDeclareOp` using `FusedLoc` approach. Codegen can use them to
create `DbgDeclareOp`.  I have added tests that checks the debug
information in mlir from and also in llvm ir.

Currently we only handle very limited types. Rest are given a place
holder type. The previous placeholder type was basic type with
`DW_ATE_address` encoding. When variables are added, it started
causing assertions in the llvm debug info generation logic for some
types. It has been changed to an interger type to prevent these issues
until we handle those types properly.

Author: abidh

flang/lib/Optimizer/CodeGen/CGOps.h renamed to flang/include/flang/Optimizer/CodeGen/CGOps.h

@@ -13,6 +13,7 @@
 #ifndef OPTIMIZER_CODEGEN_CGOPS_H
 #define OPTIMIZER_CODEGEN_CGOPS_H
 
+#include "flang/Optimizer/Dialect/FIRAttr.h"
 #include "flang/Optimizer/Dialect/FIRType.h"
 #include "mlir/Dialect/Func/IR/FuncOps.h"
 

flang/include/flang/Optimizer/CodeGen/CGOps.td

@@ -16,6 +16,8 @@
 
 include "mlir/IR/SymbolInterfaces.td"
 include "flang/Optimizer/Dialect/FIRTypes.td"
+include "flang/Optimizer/Dialect/FIRAttr.td"
+include "mlir/IR/BuiltinAttributes.td"
 
 def fircg_Dialect : Dialect {
   let name = "fircg";
@@ -202,4 +204,36 @@ def fircg_XArrayCoorOp : fircg_Op<"ext_array_coor", [AttrSizedOperandSegments]>
   }];
 }
 
+// Extended Declare operation.
+def fircg_XDeclareOp : fircg_Op<"ext_declare", [AttrSizedOperandSegments]> {
+  let summary = "for internal conversion only";
+
+  let description = [{
+    Prior to lowering to LLVM IR dialect, a DeclareOp will
+    be converted to an extended DeclareOp.
+  }];
+
+  let arguments = (ins
+    AnyRefOrBox:$memref,
+    Variadic<AnyIntegerType>:$shape,
+    Variadic<AnyIntegerType>:$shift,
+    Variadic<AnyIntegerType>:$typeparams,
+    Optional<fir_DummyScopeType>:$dummy_scope,
+    Builtin_StringAttr:$uniq_name
+  );
+  let results = (outs AnyRefOrBox);
+
+  let assemblyFormat = [{
+    $memref (`(` $shape^ `)`)? (`origin` $shift^)? (`typeparams` $typeparams^)?
+    (`dummy_scope` $dummy_scope^)?
+    attr-dict `:` functional-type(operands, results)
+  }];
+
+  let extraClassDeclaration = [{
+    // Shape is optional, but if it exists, it will be at offset 1.
+    unsigned shapeOffset() { return 1; }
+    unsigned shiftOffset() { return shapeOffset() + getShape().size(); }
+  }];
+}
+
 #endif

flang/include/flang/Optimizer/CodeGen/CGPasses.td

@@ -47,6 +47,10 @@ def CodeGenRewrite : Pass<"cg-rewrite", "mlir::ModuleOp"> {
   let dependentDialects = [
     "fir::FIROpsDialect", "fir::FIRCodeGenDialect"
   ];
+  let options = [
+    Option<"preserveDeclare", "preserve-declare", "bool", /*default=*/"false",
+           "Preserve DeclareOp during pre codegen re-write.">
+  ];
   let statistics = [
     Statistic<"numDCE", "num-dce'd", "Number of operations eliminated">
   ];

flang/include/flang/Optimizer/CodeGen/CodeGen.h

@@ -30,7 +30,8 @@ struct NameUniquer;
 
 /// Prerequiste pass for code gen. Perform intermediate rewrites to perform
 /// the code gen (to LLVM-IR dialect) conversion.
-std::unique_ptr<mlir::Pass> createFirCodeGenRewritePass();
+std::unique_ptr<mlir::Pass> createFirCodeGenRewritePass(
+    CodeGenRewriteOptions Options = CodeGenRewriteOptions{});
 
 /// FirTargetRewritePass options.
 struct TargetRewriteOptions {
@@ -88,7 +89,8 @@ void populateFIRToLLVMConversionPatterns(fir::LLVMTypeConverter &converter,
                                          fir::FIRToLLVMPassOptions &options);
 
 /// Populate the pattern set with the PreCGRewrite patterns.
-void populatePreCGRewritePatterns(mlir::RewritePatternSet &patterns);
+void populatePreCGRewritePatterns(mlir::RewritePatternSet &patterns,
+                                  bool preserveDeclare);
 
 // declarative passes
 #define GEN_PASS_REGISTRATION

flang/include/flang/Tools/CLOptions.inc

@@ -169,9 +169,11 @@ inline void addMemoryAllocationOpt(mlir::PassManager &pm) {
 }
 
 #if !defined(FLANG_EXCLUDE_CODEGEN)
-inline void addCodeGenRewritePass(mlir::PassManager &pm) {
-  addPassConditionally(
-      pm, disableCodeGenRewrite, fir::createFirCodeGenRewritePass);
+inline void addCodeGenRewritePass(mlir::PassManager &pm, bool preserveDeclare) {
+  fir::CodeGenRewriteOptions options;
+  options.preserveDeclare = preserveDeclare;
+  addPassConditionally(pm, disableCodeGenRewrite,
+      [&]() { return fir::createFirCodeGenRewritePass(options); });
 }
 
 inline void addTargetRewritePass(mlir::PassManager &pm) {
@@ -353,7 +355,8 @@ inline void createDefaultFIRCodeGenPassPipeline(mlir::PassManager &pm,
     MLIRToLLVMPassPipelineConfig config, llvm::StringRef inputFilename = {}) {
   fir::addBoxedProcedurePass(pm);
   addNestedPassToAllTopLevelOperations(pm, fir::createAbstractResultOpt);
-  fir::addCodeGenRewritePass(pm);
+  fir::addCodeGenRewritePass(
+      pm, (config.DebugInfo != llvm::codegenoptions::NoDebugInfo));
   fir::addTargetRewritePass(pm);
   fir::addExternalNameConversionPass(pm, config.Underscoring);
   fir::createDebugPasses(pm, config.DebugInfo, config.OptLevel, inputFilename);

flang/lib/Optimizer/CodeGen/CGOps.cpp

@@ -10,7 +10,7 @@
 //
 //===----------------------------------------------------------------------===//
 
-#include "CGOps.h"
+#include "flang/Optimizer/CodeGen/CGOps.h"
 #include "flang/Optimizer/Dialect/FIRDialect.h"
 #include "flang/Optimizer/Dialect/FIROps.h"
 #include "flang/Optimizer/Dialect/FIRType.h"

flang/lib/Optimizer/CodeGen/CodeGen.cpp

@@ -12,7 +12,7 @@
 
 #include "flang/Optimizer/CodeGen/CodeGen.h"
 
-#include "CGOps.h"
+#include "flang/Optimizer/CodeGen/CGOps.h"
 #include "flang/Optimizer/CodeGen/CodeGenOpenMP.h"
 #include "flang/Optimizer/CodeGen/FIROpPatterns.h"
 #include "flang/Optimizer/CodeGen/TypeConverter.h"
@@ -170,6 +170,28 @@ genAllocationScaleSize(OP op, mlir::Type ity,
   return nullptr;
 }
 
+namespace {
+struct DeclareOpConversion : public fir::FIROpConversion<fir::cg::XDeclareOp> {
+public:
+  using FIROpConversion::FIROpConversion;
+  mlir::LogicalResult
+  matchAndRewrite(fir::cg::XDeclareOp declareOp, OpAdaptor adaptor,
+                  mlir::ConversionPatternRewriter &rewriter) const override {
+    auto memRef = adaptor.getOperands()[0];
+    if (auto fusedLoc = mlir::dyn_cast<mlir::FusedLoc>(declareOp.getLoc())) {
+      if (auto varAttr =
+              mlir::dyn_cast_or_null<mlir::LLVM::DILocalVariableAttr>(
+                  fusedLoc.getMetadata())) {
+        rewriter.create<mlir::LLVM::DbgDeclareOp>(memRef.getLoc(), memRef,
+                                                  varAttr, nullptr);
+      }
+    }
+    rewriter.replaceOp(declareOp, memRef);
+    return mlir::success();
+  }
+};
+} // namespace
+
 namespace {
 /// convert to LLVM IR dialect `alloca`
 struct AllocaOpConversion : public fir::FIROpConversion<fir::AllocaOp> {
@@ -3714,19 +3736,19 @@ void fir::populateFIRToLLVMConversionPatterns(
       BoxOffsetOpConversion, BoxProcHostOpConversion, BoxRankOpConversion,
       BoxTypeCodeOpConversion, BoxTypeDescOpConversion, CallOpConversion,
       CmpcOpConversion, ConstcOpConversion, ConvertOpConversion,
-      CoordinateOpConversion, DTEntryOpConversion, DivcOpConversion,
-      EmboxOpConversion, EmboxCharOpConversion, EmboxProcOpConversion,
-      ExtractValueOpConversion, FieldIndexOpConversion, FirEndOpConversion,
-      FreeMemOpConversion, GlobalLenOpConversion, GlobalOpConversion,
-      HasValueOpConversion, InsertOnRangeOpConversion, InsertValueOpConversion,
-      IsPresentOpConversion, LenParamIndexOpConversion, LoadOpConversion,
-      MulcOpConversion, NegcOpConversion, NoReassocOpConversion,
-      SelectCaseOpConversion, SelectOpConversion, SelectRankOpConversion,
-      SelectTypeOpConversion, ShapeOpConversion, ShapeShiftOpConversion,
-      ShiftOpConversion, SliceOpConversion, StoreOpConversion,
-      StringLitOpConversion, SubcOpConversion, TypeDescOpConversion,
-      TypeInfoOpConversion, UnboxCharOpConversion, UnboxProcOpConversion,
-      UndefOpConversion, UnreachableOpConversion,
+      CoordinateOpConversion, DTEntryOpConversion, DeclareOpConversion,
+      DivcOpConversion, EmboxOpConversion, EmboxCharOpConversion,
+      EmboxProcOpConversion, ExtractValueOpConversion, FieldIndexOpConversion,
+      FirEndOpConversion, FreeMemOpConversion, GlobalLenOpConversion,
+      GlobalOpConversion, HasValueOpConversion, InsertOnRangeOpConversion,
+      InsertValueOpConversion, IsPresentOpConversion, LenParamIndexOpConversion,
+      LoadOpConversion, MulcOpConversion, NegcOpConversion,
+      NoReassocOpConversion, SelectCaseOpConversion, SelectOpConversion,
+      SelectRankOpConversion, SelectTypeOpConversion, ShapeOpConversion,
+      ShapeShiftOpConversion, ShiftOpConversion, SliceOpConversion,
+      StoreOpConversion, StringLitOpConversion, SubcOpConversion,
+      TypeDescOpConversion, TypeInfoOpConversion, UnboxCharOpConversion,
+      UnboxProcOpConversion, UndefOpConversion, UnreachableOpConversion,
       UnrealizedConversionCastOpConversion, XArrayCoorOpConversion,
       XEmboxOpConversion, XReboxOpConversion, ZeroOpConversion>(converter,
                                                                 options);

flang/lib/Optimizer/CodeGen/PreCGRewrite.cpp

@@ -12,8 +12,8 @@
 
 #include "flang/Optimizer/CodeGen/CodeGen.h"
 
-#include "CGOps.h"
 #include "flang/Optimizer/Builder/Todo.h" // remove when TODO's are done
+#include "flang/Optimizer/CodeGen/CGOps.h"
 #include "flang/Optimizer/Dialect/FIRDialect.h"
 #include "flang/Optimizer/Dialect/FIROps.h"
 #include "flang/Optimizer/Dialect/FIRType.h"
@@ -270,13 +270,43 @@ class ArrayCoorConversion : public mlir::OpRewritePattern<fir::ArrayCoorOp> {
 };
 
 class DeclareOpConversion : public mlir::OpRewritePattern<fir::DeclareOp> {
+  bool preserveDeclare;
+
 public:
   using OpRewritePattern::OpRewritePattern;
+  DeclareOpConversion(mlir::MLIRContext *ctx, bool preserveDecl)
+      : OpRewritePattern(ctx), preserveDeclare(preserveDecl) {}
 
   mlir::LogicalResult
   matchAndRewrite(fir::DeclareOp declareOp,
                   mlir::PatternRewriter &rewriter) const override {
-    rewriter.replaceOp(declareOp, declareOp.getMemref());
+    if (!preserveDeclare) {
+      rewriter.replaceOp(declareOp, declareOp.getMemref());
+      return mlir::success();
+    }
+    auto loc = declareOp.getLoc();
+    llvm::SmallVector<mlir::Value> shapeOpers;
+    llvm::SmallVector<mlir::Value> shiftOpers;
+    if (auto shapeVal = declareOp.getShape()) {
+      if (auto shapeOp = mlir::dyn_cast<fir::ShapeOp>(shapeVal.getDefiningOp()))
+        populateShape(shapeOpers, shapeOp);
+      else if (auto shiftOp =
+                   mlir::dyn_cast<fir::ShapeShiftOp>(shapeVal.getDefiningOp()))
+        populateShapeAndShift(shapeOpers, shiftOpers, shiftOp);
+      else if (auto shiftOp =
+                   mlir::dyn_cast<fir::ShiftOp>(shapeVal.getDefiningOp()))
+        populateShift(shiftOpers, shiftOp);
+      else
+        return mlir::failure();
+    }
+    // FIXME: Add FortranAttrs and CudaAttrs
+    auto xDeclOp = rewriter.create<fir::cg::XDeclareOp>(
+        loc, declareOp.getType(), declareOp.getMemref(), shapeOpers, shiftOpers,
+        declareOp.getTypeparams(), declareOp.getDummyScope(),
+        declareOp.getUniqName());
+    LLVM_DEBUG(llvm::dbgs()
+               << "rewriting " << declareOp << " to " << xDeclOp << '\n');
+    rewriter.replaceOp(declareOp, xDeclOp.getOperation()->getResults());
     return mlir::success();
   }
 };
@@ -297,6 +327,7 @@ class DummyScopeOpConversion
 
 class CodeGenRewrite : public fir::impl::CodeGenRewriteBase<CodeGenRewrite> {
 public:
+  CodeGenRewrite(fir::CodeGenRewriteOptions opts) : Base(opts) {}
   void runOnOperation() override final {
     mlir::ModuleOp mod = getOperation();
 
@@ -314,7 +345,7 @@ class CodeGenRewrite : public fir::impl::CodeGenRewriteBase<CodeGenRewrite> {
                    mlir::cast<fir::BaseBoxType>(embox.getType()).getEleTy()));
     });
     mlir::RewritePatternSet patterns(&context);
-    fir::populatePreCGRewritePatterns(patterns);
+    fir::populatePreCGRewritePatterns(patterns, preserveDeclare);
     if (mlir::failed(
             mlir::applyPartialConversion(mod, target, std::move(patterns)))) {
       mlir::emitError(mlir::UnknownLoc::get(&context),
@@ -330,12 +361,14 @@ class CodeGenRewrite : public fir::impl::CodeGenRewriteBase<CodeGenRewrite> {
 
 } // namespace
 
-std::unique_ptr<mlir::Pass> fir::createFirCodeGenRewritePass() {
-  return std::make_unique<CodeGenRewrite>();
+std::unique_ptr<mlir::Pass>
+fir::createFirCodeGenRewritePass(fir::CodeGenRewriteOptions Options) {
+  return std::make_unique<CodeGenRewrite>(Options);
 }
 
-void fir::populatePreCGRewritePatterns(mlir::RewritePatternSet &patterns) {
+void fir::populatePreCGRewritePatterns(mlir::RewritePatternSet &patterns,
+                                       bool preserveDeclare) {
   patterns.insert<EmboxConversion, ArrayCoorConversion, ReboxConversion,
-                  DeclareOpConversion, DummyScopeOpConversion>(
-      patterns.getContext());
+                  DummyScopeOpConversion>(patterns.getContext());
+  patterns.add<DeclareOpConversion>(patterns.getContext(), preserveDeclare);
 }

flang/lib/Optimizer/Transforms/AddDebugInfo.cpp

@@ -15,6 +15,7 @@
 #include "flang/Common/Version.h"
 #include "flang/Optimizer/Builder/FIRBuilder.h"
 #include "flang/Optimizer/Builder/Todo.h"
+#include "flang/Optimizer/CodeGen/CGOps.h"
 #include "flang/Optimizer/Dialect/FIRDialect.h"
 #include "flang/Optimizer/Dialect/FIROps.h"
 #include "flang/Optimizer/Dialect/FIRType.h"
@@ -45,13 +46,59 @@ namespace fir {
 namespace {
 
 class AddDebugInfoPass : public fir::impl::AddDebugInfoBase<AddDebugInfoPass> {
+  void handleDeclareOp(fir::cg::XDeclareOp declOp,
+                       mlir::LLVM::DIFileAttr fileAttr,
+                       mlir::LLVM::DIScopeAttr scopeAttr,
+                       fir::DebugTypeGenerator &typeGen);
+
 public:
   AddDebugInfoPass(fir::AddDebugInfoOptions options) : Base(options) {}
   void runOnOperation() override;
 };
 
+static uint32_t getLineFromLoc(mlir::Location loc) {
+  uint32_t line = 1;
+  if (auto fileLoc = mlir::dyn_cast<mlir::FileLineColLoc>(loc))
+    line = fileLoc.getLine();
+  return line;
+}
+
 } // namespace
 
+void AddDebugInfoPass::handleDeclareOp(fir::cg::XDeclareOp declOp,
+                                       mlir::LLVM::DIFileAttr fileAttr,
+                                       mlir::LLVM::DIScopeAttr scopeAttr,
+                                       fir::DebugTypeGenerator &typeGen) {
+  mlir::MLIRContext *context = &getContext();
+  mlir::OpBuilder builder(context);
+  auto result = fir::NameUniquer::deconstruct(declOp.getUniqName());
+
+  if (result.first != fir::NameUniquer::NameKind::VARIABLE)
+    return;
+
+  // Only accept local variables.
+  if (result.second.procs.empty())
+    return;
+
+  // FIXME: There may be cases where an argument is processed a bit before
+  // DeclareOp is generated. In that case, DeclareOp may point to an
+  // intermediate op and not to BlockArgument. We need to find those cases and
+  // walk the chain to get to the actual argument.
+
+  unsigned argNo = 0;
+  if (auto Arg = llvm::dyn_cast<mlir::BlockArgument>(declOp.getMemref()))
+    argNo = Arg.getArgNumber() + 1;
+
+  auto tyAttr = typeGen.convertType(fir::unwrapRefType(declOp.getType()),
+                                    fileAttr, scopeAttr, declOp.getLoc());
+
+  auto localVarAttr = mlir::LLVM::DILocalVariableAttr::get(
+      context, scopeAttr, mlir::StringAttr::get(context, result.second.name),
+      fileAttr, getLineFromLoc(declOp.getLoc()), argNo, /* alignInBits*/ 0,
+      tyAttr);
+  declOp->setLoc(builder.getFusedLoc({declOp->getLoc()}, localVarAttr));
+}
+
 void AddDebugInfoPass::runOnOperation() {
   mlir::ModuleOp module = getOperation();
   mlir::MLIRContext *context = &getContext();
@@ -144,14 +191,15 @@ void AddDebugInfoPass::runOnOperation() {
       subprogramFlags =
           subprogramFlags | mlir::LLVM::DISubprogramFlags::Definition;
     }
-    unsigned line = 1;
-    if (auto funcLoc = mlir::dyn_cast<mlir::FileLineColLoc>(l))
-      line = funcLoc.getLine();
-
+    unsigned line = getLineFromLoc(l);
     auto spAttr = mlir::LLVM::DISubprogramAttr::get(
         context, id, compilationUnit, fileAttr, funcName, fullName,
         funcFileAttr, line, line, subprogramFlags, subTypeAttr);
     funcOp->setLoc(builder.getFusedLoc({funcOp->getLoc()}, spAttr));
+
+    funcOp.walk([&](fir::cg::XDeclareOp declOp) {
+      handleDeclareOp(declOp, fileAttr, spAttr, typeGen);
+    });
   });
 }
 

flang/lib/Optimizer/Transforms/DebugTypeGenerator.cpp

@@ -24,11 +24,6 @@ DebugTypeGenerator::DebugTypeGenerator(mlir::ModuleOp m)
   LLVM_DEBUG(llvm::dbgs() << "DITypeAttr generator\n");
 }
 
-static mlir::LLVM::DITypeAttr genPlaceholderType(mlir::MLIRContext *context) {
-  return mlir::LLVM::DIBasicTypeAttr::get(
-      context, llvm::dwarf::DW_TAG_base_type, "void", 32, 1);
-}
-
 static mlir::LLVM::DITypeAttr genBasicType(mlir::MLIRContext *context,
                                            mlir::StringAttr name,
                                            unsigned bitSize,
@@ -37,6 +32,11 @@ static mlir::LLVM::DITypeAttr genBasicType(mlir::MLIRContext *context,
       context, llvm::dwarf::DW_TAG_base_type, name, bitSize, decoding);
 }
 
+static mlir::LLVM::DITypeAttr genPlaceholderType(mlir::MLIRContext *context) {
+  return genBasicType(context, mlir::StringAttr::get(context, "integer"), 32,
+                      llvm::dwarf::DW_ATE_signed);
+}
+
 mlir::LLVM::DITypeAttr
 DebugTypeGenerator::convertType(mlir::Type Ty, mlir::LLVM::DIFileAttr fileAttr,
                                 mlir::LLVM::DIScopeAttr scope,